Better Know a Data Scientist
Michel Manago is CEO of Kiolis, a Paris-based startup that develops software for personalization and recommendation. I interviewed Michel about the MyCoachNutrition project, which applies case-based reasoning and collaborative filtering technologies to guide subscribers to eat and exercise well. It strikes me that the project combines 3 of Michel’s keenest interests: algorithms, starting companies, and food. Given his achievements in these 3 disciplines, Michel is always at the top of my list for dinner companions.
- How would you describe the current status of MyCoachNutrition?
MyCoachNutrition is in alpha. The broad commercial launch, through a distributor that will promote us to 2.5 million French families, is scheduled for August 2015. Beta is scheduled in May with a limited commercial launch in June.
- What challenge surprised you in getting to this point?
The biggest challenge was to create a value proposition that is sufficiently attractive for our users. Some of our users are used to getting free content, such as cooking recipes, on most web site. However, they do not necessarily trust this information. Content has to be free or fantastic as one of my clients used to say. We are not in the business of providing free content, therefore our content must be fantastic.
Our content has previously been published by a professional publishers, but our dietician and our pharmacologist have reviewed every single bit of it in order to check the accuracy of our nutritional facts. We originally assumed that our content would be fantastic because it had been published by a professional publisher but, unfortunately, we quickly realized that you do not publish in the same way on the web as you do in books. The content was not unified because each book lives its own life and has its own constraints. When we combined the material from the different books (we have access to over 10,000 published cooking recipes for example), we quickly noticed that the metadata to classify this content was different in each book. Even basic parameters, such as “how easy it is to cook this recipe”, “what is the cost of the recipe”, “how long it takes” were not consistently provided for every recipes in every book. Not to mention other more advanced criteria such as “the country”, “the theme” etc. So we had to unify this content and enhance it so that :
A). our users can search based on these criteria.
B). our recommendation engine can use these criteria to create personalized recommendations based on what we know about our users (cooking skills, budget, available time, type of food they like etc.)
Because we knew our users wanted to increase their skills, we also develop our own video content to teach cooking techniques. We reviewed many web sites with video and designed our own guidelines. For example, a video should not be more than 3 minutes at most. It should be divided in steps that match the cooking steps etc. We ended up developing our own player so the user can view the steps in text form and pause in between steps. A video’s function is not merely to be nice, but it has to be almost like an individual e-learning project. What do we want to teach through this video? How effective are we? Can someone, who does not have the skills of our chef, easily reproduce what we show on the video?
- How did you achieve your value proposition?
We tested the concept early on some focus groups. We ran tests with 4 focus groups of 5 people for 2 hours each. Originally, we assumed that our main audience would be people who want to lose weight or people who eat specific types of food (ex: vegetarians, or people with allergies). We then discovered, during these early tests, that these were not the most promising audience for myCoachNutrition. We now focus on families and on people who want to feel in good health. We also found out that while our users appreciated being told how to exercise and what to eat, they also wanted to learn and to understand why we made the recommendations. We switch from a system that would propose ready-made meal plans to a system that is totally interactive. The user receives assistance but only if and when he/she requests it. It’s almost like an interactive game where one learns how to balance his/her own meals and learn what to order do when he/she is not at home to prepare his/her own meals.
- What is the biggest challenge you expected, and how did you address it?
The biggest challenge was to develop a system that is at the same time entertaining and fun to use, but that has the credibility of a proven scientific approach. During a three year joint research project (the fiora e-health project www.fiora.pro), we gathered the know-how of the best French specialists in nutrition and health and combined it with the big data expertise of IT experts from the public and private sectors. The next challenge was to put the technology to use and make this advanced know-how accessible to a wide audience as well as gain visibility so that we will generate revenues through subscriptions. We achieved this by making a deal with a French distributor who gives us visibility with their 2.5 million subscribers. Finally, health-data privacy is a major challenge and we make sure that our users understand that their data is secure and remain private. We do not sell their data to third parties, nor do we generate revenues through advertising. The guarantee that the data from our users will remain safe with us has a cost for our users (the monthly subscription fee), but this builds trust and trust is key to our business.